GRBM Spectra.

Since the 128 s spectra are acquired independently from the ratemeters, regardless of possible trigger times, their 128 s time intervals available for the two flares are the following: taking the trigger as zero-time, for the 1998 flare there are three 128 s spectra: [-59.6 s, +68.4s] (A), [+68.4 s, +196.4 s] (B), and [+196.4 s, +324.4 s] (C); in the 2001 flare case, the situation is easier, owing to the shorter duration: [-95.4 s, +32.6 s] (A) and [+32.6 s, +160.6 s] (B). The 1998 flare slice (A) is strongly affected by 16-bit counter recycling, so it needs a careful analysis; here only the slice (B) spectrum is considered, in addition to the slice (A) spectrum for the 2001 flare, which nearly includes the whole flare, apart from the last separated pulse around $t \sim 40$ s. The fitted spectra and correspondent values of the best-fitting parameters are shown in figg.  and reported within table . As for the 1998 flare spectrum (B), by comparing our fit results with black body plus broken power law (see table ), with those reported by [Feroci et al., 1999a], that fit the data with black body plus single power law ( $kT = 15.5 \pm 1.0$ keV and $\alpha = 4.5 \pm 0.1$, $\chi^2 = 1.424$), the different results might be due to the different response matrices used: [Feroci et al., 1999a], added in the fits a 10% systematic error to account for the bad knowledge of the GRBM response for that local direction ; in our analysis no systematic error has been considered, thanks to the good knowledge of the GRBM response function, and the fit has been improved.

To some extent, the goodness of this fit is confirmed also by the fit based on the same law (black body plus broken power law), when applied to the 2001 flare spectrum: despite the local direction of the source is $\sim 49\rm ^{\circ}$ far from the 1998 flare one (table ), and, therefore, the GRBM effective area is significantly different, the parameters found by applying the same kind of spectrum (black body plus broken power law) are slightly different from the 1998 case (table ), except for the BB temperature, that is $\sim 14.4 \pm 0.4$ keV against $kT = 18.0 \pm 0.6$ keV (1998); this is consistent with the higher energy released during the 1998 flare than in the 2001 event.